Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PPGN: Physics-Preserved Graph Networks for Real-Time Fault Location in Distribution Systems with Limited Observation and Labels

Wenting Li, Deepjyoti Deka

TL;DR

The paper tackles real-time fault location in distribution networks under limited observability and scarce labels by introducing PPGN, a two-stage physics-preserved graph network. Stage I builds a graph embedding with an adjustable adjacency $A$ to reflect grid physics and sparse observability, while Stage II propagates labels through a correlation graph with adjacency $B$ derived from Stage I embeddings, grounded by a random-walk interpretation. The approach yields superior accuracy and robustness on IEEE 123-node and 37-node feeders, notably at low label rates and under load/topology variations, outperforming three baselines. The work highlights the value of integrating physical grid relationships into graph learning for critical monitoring tasks and points to future PMU placement optimization to further enhance performance.

Abstract

Electrical faults may trigger blackouts or wildfires without timely monitoring and control strategy. Traditional solutions for locating faults in distribution systems are not real-time when network observability is low, while novel black-box machine learning methods are vulnerable to stochastic environments. We propose a novel Physics-Preserved Graph Network (PPGN) architecture to accurately locate faults at the node level with limited observability and labeled training data. PPGN has a unique two-stage graph neural network architecture. The first stage learns the graph embedding to represent the entire network using a few measured nodes. The second stage finds relations between the labeled and unlabeled data samples to further improve the location accuracy. We explain the benefits of the two-stage graph configuration through a random walk equivalence. We numerically validate the proposed method in the IEEE 123-node and 37-node test feeders, demonstrating the superior performance over three baseline classifiers when labeled training data is limited, and loads and topology are allowed to vary.

PPGN: Physics-Preserved Graph Networks for Real-Time Fault Location in Distribution Systems with Limited Observation and Labels

TL;DR

The paper tackles real-time fault location in distribution networks under limited observability and scarce labels by introducing PPGN, a two-stage physics-preserved graph network. Stage I builds a graph embedding with an adjustable adjacency to reflect grid physics and sparse observability, while Stage II propagates labels through a correlation graph with adjacency derived from Stage I embeddings, grounded by a random-walk interpretation. The approach yields superior accuracy and robustness on IEEE 123-node and 37-node feeders, notably at low label rates and under load/topology variations, outperforming three baselines. The work highlights the value of integrating physical grid relationships into graph learning for critical monitoring tasks and points to future PMU placement optimization to further enhance performance.

Abstract

Electrical faults may trigger blackouts or wildfires without timely monitoring and control strategy. Traditional solutions for locating faults in distribution systems are not real-time when network observability is low, while novel black-box machine learning methods are vulnerable to stochastic environments. We propose a novel Physics-Preserved Graph Network (PPGN) architecture to accurately locate faults at the node level with limited observability and labeled training data. PPGN has a unique two-stage graph neural network architecture. The first stage learns the graph embedding to represent the entire network using a few measured nodes. The second stage finds relations between the labeled and unlabeled data samples to further improve the location accuracy. We explain the benefits of the two-stage graph configuration through a random walk equivalence. We numerically validate the proposed method in the IEEE 123-node and 37-node test feeders, demonstrating the superior performance over three baseline classifiers when labeled training data is limited, and loads and topology are allowed to vary.

Paper Structure

This paper contains 18 sections, 17 equations, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Physics of Fault Currents and Problem Formulation
  3. Sparsity of Fault Currents
  4. Problem Formulation
  5. Proposed Graph Learning framework
  6. Stage I: Graph Embedding Learning
  7. Stage II: Label Propagation
  8. Theoretical Interpretations
  9. Construct $A$ to Improve Visibility
  10. Construct $B$ to Enhance the Exact Prediction Probability
  11. Numerical Experiments
  12. Implementation Details
  13. Performance Comparison
  14. Two-stage Location Performance
  15. Performance at Different Stages When Label Rates are Low
  16. ...and 3 more sections

Figures (4)

  • Figure 1: The IEEE 123-node test feeder
  • Figure 2: The structure of PPGN
  • Figure 3: Example: Filled nodes are measured, and void nodes are unmeasured. When $k_{\text{I}} = 2, K=1$, node $i$ only has one connected node $j_1$ in (a), but $i$ has three neighbors $j_k,k=1,2,3$ in (b) defined by our $A$, which ensures that there are some paths from the unmeasured nodes to the measured nodes.
  • Figure 4: LAR at Different Label Rates